The model proposes.
Statistics dispose.
Your own experiment confirms.

Write or say what happened. Listening pulls out sleep, mood, exercise, and the people in your life, then looks for patterns in your own data.

Start writing Free to start · English and Chinese

No streaks
No scores
No wearable
No integrations

Tuesday 11:04 pm

slept badly, maybe 5 hours

played soccer for an hour, absolutely wrecked. dinner with Marco after

Soccer — an hour, or closer to 90 minutes?

60m 90m

Extracted from these words

sleep 5h exercise 60m person: Marco

Tap any chip to see the exact words it came from.

How it works

01

You write it down

Type or speak, in whatever words you'd use. No fields, no dropdowns. English, Chinese, or both in one sentence.
02

It reads, and asks when unsure

Sleep, mood, exercise, people. Ambiguity becomes a one-tap question, never a guess.
03

It proposes, then tests

The model writes hypotheses. Statistics decide which ones hold up.

Why most tracking apps lie to you

Not on purpose. They lie in three structural ways.

1

A model that reads sixty days of your life will always find something

Ask a language model for insights and you get insights. It has no way to say "there is nothing here." Given enough of your text it will produce a confident paragraph about how Tuesdays drain you, and it will be well written, and it will be astrology.

Here the model is plumbing. It proposes candidates; a statistical layer decides which survive — effect size, how many days contribute, how many hypotheses were tested. Most die there, and you never see them.

2

"Not mentioned" is not "didn't happen"

A day you didn't write about is a day with no data. Most apps quietly file it as a zero, and every average after that is wrong in the same direction.

Listening keeps the gap. A statistic you can't trust doesn't get drawn.

Sleep, last nine days

Three days you didn't write. Shown as gaps, left out of the average, counted in the sample size you see.

3

Rewarding you for good numbers corrupts the numbers

A streak on a good value is a reason to log a good value. Once the record is something you can win at, it stops describing your life and starts describing the game.

So: no streaks, no scores, no badges. Colour never evaluates. The only thing the app wants from you is an accurate sentence.

Three rungs, and only the top one gives advice

Every finding carries its rung, so you always know how much weight it can hold.

Rung 1
Observed
A pattern the model noticed.

Might be coincidence. Says nothing about cause, suggests nothing.
Rung 2
Supported
It survived the statistics.

Enough days, a real effect size, and it held up after accounting for everything else tested. Still correlation. Still no advice.
Rung 3
Validated
You ran the experiment.

A randomized n-of-1 trial: the app assigns the condition day by day, you live it, then it reads the result. Only findings here may recommend anything.

No wearable. No calendar. No integrations.

Nothing to pair, nothing to authorize, nothing that breaks when a vendor changes an API. The data comes from you.

Two check-ins a day cover sleep and mood, about ten seconds each. Everything else lands on the factors board, pre-filled from what you wrote.

Today's factors

from your entries

Caffeine after 2pm ✓ confirmed
Alcohol ✗ didn't happen
Screen time in bed — no data

Three states, and the third one is honest. A blank is never read as a no.

Your words are the asset

Entries are written once and never rewritten. Everything derived from them is a cache that can be thrown away and rebuilt.

Voice is transcribed on your iPhone by a model the app downloads for you. Recordings are never uploaded; the transcript does go to the server for extraction.

Immutable. Corrections are added, not painted over.
Exportable. JSON and markdown, any time. No paywall on your own words.
Deletable. One value, one entry, or the whole account — from inside the app.

A notebook that argues with itself before it argues with you.